Sixth European Conference on Speech Communication and Technology

Budapest, Hungary
September 5-9, 1999

Learning of Domain Dependent Knowledge in Semantic Networks

F. Deinzer, J. Fischer, U. Ahlrichs, Elmar Nöth

Chair for Pattern Recognition, University of Erlangen-Nuremberg, Erlangen, Germany

For an eficient linguistic analysis of spoken queries alot of domain specific knowledge is needed and usuallyhas to be entered manually into the knowledge baseof each domain. This makes the adaption of dialoguesystems which base on explicit knowledge representation to new domains a very costly procedure. We usea frequency based statistical method combined withgeneral hidden markov models in order to learn do-main specific knowledge within a semantic networkformalism. As a framework we use a dialogue systemfor German train timetable information. By means ofexperiments we show that our statistical approach isnot only able to reach, but even outperforms previousresults with manually entered restrictions.

Full Paper (PDF)   Gnu-Zipped Postscript

Bibliographic reference.  Deinzer, F. / Fischer, J. / Ahlrichs, U. / Nöth, Elmar (1999): "Learning of domain dependent knowledge in semantic networks", In EUROSPEECH'99, 1987-1990.